Three-dimensional Reconstruction of Rough Surfaces Using Epipolar Geometry

Abstract

The computer vision study objective is to automatically or semi-automatically extract objects useful information to generate 3D models through the use of image processing techniques. This project will recover the three dimensional reconstruction of rough surfaces to understand the interfacial interaction phenomena that can be found in corrosion and tribology (friction, wear and lubrification study). The surface roughness is normally recovered using a surface contact gauge, or laser-based gauges along the line. In the surface contact gauge, the plane contact occurs between the plane highest peak, that doesn't correspond to the peak along the line. To evaluate the real contact between the planes a 3D reconstruction of this rough surface is required and will be achieved using a microscope and a stereo camera. In the 3D reconstruction a microscope and a stereo camera will be used and different techniques will be used for this process: 1. automatic using SIFT, ICP, RANSAC algorithms in the images; 2. automatic pointing a laser on the rough surface and capturing a stereo movie to determine the keypoints; and 3. automatic using a camera calibration, SIFT and ICP. The project scope is the development and test of the above methods, as well as the errors of each method. This project will be developed with the "Laboratório de Fenômenos de Superfície do Departamento de Engenharia Mecânica" of the EPUSP. Once 3D reconstruction is finished, the contact points between surfaces will be recovered. During the postdoctoral (CNPQ scholarship) and the doctorate (japanese government scholarship) the candidate developed some applications that will be useful in this project. The following applications were developed: 1. keypoints determination (SIFT); 2. keypoints mapping (ICP); 3. epipolar geometry recovery using 7 points matches method; 4. epipolar geometry validation (improved RANSAC) using the topology preservation of the object; and 5. 3D points coordinate recovery.